US10062144B2 - Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition - Google Patents

Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition Download PDF

Info

Publication number
US10062144B2
US10062144B2 US15642836 US201715642836A US10062144B2 US 10062144 B2 US10062144 B2 US 10062144B2 US 15642836 US15642836 US 15642836 US 201715642836 A US201715642836 A US 201715642836A US 10062144 B2 US10062144 B2 US 10062144B2
Authority
US
Grant status
Grant
Patent type
Prior art keywords
invariants
method
visual image
invariant
recited
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active, expires
Application number
US15642836
Other versions
US20180060999A1 (en )
Inventor
Alex Simon Blaivas
Original Assignee
Alex Simon Blaivas
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Grant date

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00Geometric image transformation in the plane of the image, e.g. from bit-mapped to bit-mapped creating a different image
    • G06T3/005Geometric image transformation in the plane of the image, e.g. from bit-mapped to bit-mapped creating a different image for projecting an image on a non-planar surface, e.g. a geodetic screen
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/36Image preprocessing, i.e. processing the image information without deciding about the identity of the image
    • G06K9/46Extraction of features or characteristics of the image
    • G06K9/52Extraction of features or characteristics of the image by deriving mathematical or geometrical properties from the whole image
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6201Matching; Proximity measures
    • G06K9/6202Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
    • G06K9/6203Shifting or otherwise transforming the patterns to accommodate for positional errors

Abstract

A spherical harmonic is defined which is an operationally optimal small finite subset of the infinite number of spherical harmonics allowed to exist mathematically. The composition of the subset differs depending on its position on virtual hemisphere. The subsets are further divided into small spherical tesserae whose dimensions vary depending on the distance from the hemispherical center. The images of the outside visual scenes are projected on the flat surface of the webcam and from there are read and recalculated programmatically as if the images have been projected on the hemisphere, rotational invariants are then computed in the smallest tesserae using numerical integration, and then invariants from neighboring tesserae are added to compute the rotational invariant of their union. Every computed invariant is checked with the library and stored there if there is no match. The rotational invariants are solely used for visual recognition and classification and operational decision making.

Description

RELATED APPLICATIONS

This application is a divisional of U.S. patent application Ser. No. 15/251,751, filed Aug. 30, 2016, which is incorporated herein in its entirety, and is related to U.S. Pat. No. 8,670,621 (Blaivas), entitled USING INVARIANTS OF SPHERICAL HARMONICS TO ROTATIONAL AND TRANSLATIONAL TRANSFORMATIONS FOR IMAGE RECOGNITION IN INDUSTRIAL AND SECURITY APPLICATIONS, and incorporated herein in its entirety by this reference.

FIELD OF THE INVENTION

Embodiments of the present invention generally concern the construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition. More particularly, at least some embodiments of the invention relate to systems, hardware, computer-readable media, and methods directed to the implementation and use of electronic visual recognition, and the use of such electronic visual recognition in various systems, devices, and applications.

BACKGROUND

In order for a visual recognition system to operate effectively, it must be able to generate and/or access reliable data concerning the surrounding environment. As well, a visual recognition system must be able to respond and adapt to changes, which can occur rapidly and unexpectedly, in the surrounding environment. However, typical visual recognition systems may not be well equipped to perform as needed.

Accordingly, it would be useful to provide systems and methods that can generate invariants of spherical harmonics which can be employed by a visual recognition system. As well, it would be useful to provide systems and methods that are capable of generating invariants of spherical harmonics from scratch. It would also be useful to provide invariants of spherical harmonics, such as in a library, to other systems and devices so that those systems and devices can employ visual recognition without the need for any training or conditioning. Finally, it would be useful to be able to perform visual recognition processes using only a limited set of spherical harmonics to generate invariants.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to describe the manner in which at least some aspects of this disclosure can be obtained, a more particular and detailed description will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. Understanding that these drawings depict only example embodiments of the invention and are not therefore to be considered to be limiting of its scope, embodiments of the invention will be described and explained with additional specificity and detail through the use of the accompanying drawings, in which:

FIG. 1 is a flow diagram disclosing internalized computational environment shaped to perform all operations for efficient image recognition system;

FIG. 2 discloses aspects of a relation between a physical input photosensitive plane (e.g. a webcam) and a virtual hemispherical projection surface;

FIG. 3 discloses aspects relating to variability in spherical harmonics;

FIG. 4 discloses aspects relating to variability in Legendre polynomials;

FIG. 5 discloses aspects relating to variability in expansion coefficients of an example visual function;

FIG. 6 discloses an example configuration of various functional zones on a hemispherical surface;

FIG. 7 discloses an example configuration of various circular zones on a virtual 2-hemisphere with an example of a part of a spherical harmonics function divided into a few small functional rectangles;

FIG. 8 discloses an example complex exponent with two values of index m;

FIG. 9 discloses three fully normalized Associated Legendre Polynomials (ALP) with the same degree and different orders;

FIG. 10 discloses a collection of six fully normalized Associated Legendre Polynomials which are paired for comparison;

FIG. 11 discloses a collection of six fully normalized Associated Legendre Polynomials which are also paired for comparison;

FIG. 12 discloses a further collection of Associated Legendre Polynomials for comparison;

FIG. 13 discloses two independent sets of spherical harmonics on a virtual 2-hemisphere with North Poles rotated with respect to each other; and

FIG. 14 discloses bipolar spherical harmonics as a tool for binocular vision.

DETAILED DESCRIPTION OF SOME EXAMPLE EMBODIMENTS

Embodiments of the present invention generally concern the construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition. More particularly, at least some embodiments of the invention relate to systems, hardware, computer-readable media, and methods directed to the implementation and use of electronic visual recognition, and the use of such electronic visual recognition in various systems, devices, and applications. Such visual recognition functionality may include recognition of one or more physical elements of static and/or dynamic operating environments, as well as creating a comprehensive description of the whole visual scenes with highlighting especially important and/or dangerous security configurations, and can be employed in a wide variety of applications, one example of which is fully-autonomous, or semi-autonomous, control of land-based vehicles or amphibious land/sea vehicles, robots in fire or explosion prone places, and war zones, although the scope of the invention is not limited to this example application and the disclosed technology is suitable for a variety of other applications as well, such as the examples disclosed in the application identified in the ‘Related Applications’ section of this application.

In more detail, a system can be created and implemented that is configured to generate invariants based on input concerning one or more physical scenes or physical environments. One example of such a system includes one or more cameras, such as a webcam or a video PC camera, for example. The camera or cameras can snapshot two dimensional, and/or three dimensional, information regarding one or more static and/or dynamic physical environments. Regardless of the type of physical environment involved, the one or more cameras can snapshot information and then upload the recorded data for processing, such as, for example, by a hardware processor and executable instructions of a computing device or a server. More particularly, the uploaded camera data can be used for generation of invariants relating to the physical environment whose attributes were recorded.

The image data received from the cameras can be virtually projected onto a virtual Hemisphere programmatically using the obtained digital input from the camera. A real world physical analog to the Hemisphere is a retina of an eye. In at least some embodiments, only a single set of spherical harmonics is needed to generate a set of invariants.

A coordinates library correlates the subdivision of the set of spherical harmonics with a respective set of generally rectangularly shaped elements of a surface of a Hemisphere. The size of the element varies according to its location on the sphere. Invariants, and specifically rotational invariants, can now be computed for each of the elements.

After the invariants are generated for the smallest visual elements (see FIG. 1 & FIG. 7), invariants can then be generated for larger elements by adding invariants of neighboring elements together. The generated invariants can then be stored. In general, use of the term ‘invariant’ reflects the fact that the invariants represent one or more details of a visual scene that remain the same, regardless of where the detail is located on the hemisphere. Multiple sets of invariants, each corresponding to a different timeframe, can be generated. Each timeframe or time interval can correspond to a respective change in the visual scene, such as when an object within the scene is moving.

Over time, some invariants will appear relatively more often than others and the frequency with which the invariants appear can be recorded. Invariants that appear on a recurring basis are considered as indicating the true representation of an object in the physical visual world, while invariants that do not regularly appear may be considered to be artifacts not representative of the object and which can thus be discarded after the passage of a certain amount of time.

The system will also provide for generalization of invariants illustrated by this example. At the initial stage of the operation, the vision system may calculate and store invariants describing, for example, 100 various individual automobiles. Although those individual invariants are important and must be kept, the system will also generalize features common to all of them and store an image of the automobile for subsequent classification and decision making.

After the library of invariants has been reduced down to a grouping of invariants that are considered to form a true representation of an object, or objects, those remaining invariants can then be used by a visual recognition system, examples of which are disclosed herein.

Advantageously then, at least some embodiments of the invention provide for creation of invariants from scratch, that is, from visual data gathered from an actual physical operating environment. As well, the invariants can be refined over time so that artifacts are removed, thus leaving a true representation of an object or environment. Moreover, the refined set of invariants can be uploaded to a device that consists of, or comprises, a visual recognition system in its native, original state with zero number of invariants in it, or alternatively to a system that has been exposed to different operating environments and which developed visual invariants pertaining to this different environment. Because the refined set of invariants fully represents, for example, an operating environment, or at least represents the operating environment at minimal acceptable level of detail, a device employing the invariants can quickly analyze an operating environment by extracting invariant information from the viewed environment and comparing the extracted information with the pre-loaded refined set of invariants, and then taking action accordingly.

As will be apparent from this disclosure, embodiments of the invention can be employed in a variety of specific applications involving visual recognition. Moreover, it will further be apparent from this disclosure that example embodiments of the invention are effective in solving various problems in the field of visual recognition, and/or in improving currently existing visual recognition systems, methods, and devices.

A. DECISION MAKING

It should be understood that the disclosed system of visual recognition operating on the basis of the most economical representation of elements of the visual world via rotational invariants can be easily configured to operate as a decision maker, such as, for example, for a warning system of appropriate kind.

B. EXAMPLE OPERATING ENVIRONMENTS

In general, example embodiments of the invention can be employed in connection with data that corresponds to or describes static and/or dynamic physical environments. As employed herein, a static physical environment is one in which few or no visually perceptible changes occur over a defined period of time. A static physical environment can also be an environment in which one or more visually perceptible changes occur over a defined period of time, but those visually perceptible changes are ignored in the generation of invariants for that physical environment. For example, an airplane passing overhead is a transient event and may have no bearing on the physical environment for which invariants are to be generated. Thus, the passage of the airplane can be ignored, in some instances at least. Or alternatively such a relatively uncommon event like the passing of an airplane could represent the expected and desired occurrence that might trigger a decision to alert either a human observer, network or simply register a statistically significant happening.

On the other hand, a dynamic physical environment is one in which one or more visually perceptible changes occur over a defined period of time. One, some, or all, of the visually perceptible changes in a dynamic physical environment can be taken into account when generating invariants. An example of such a visually perceptible change is an event that occurs repeatedly over time in a relatively predictable manner, such as the raising and lowering of a bridge for example.

C. EXAMPLE PROCESSES AND METHODS

In general, any of the disclosed methods and processes can take the form of instructions carried on a non-transitory storage medium and executable by, and/or at the direction of, one or more hardware processors and/or other computer hardware. Such non-transitory storage media and the processors or other hardware that execute the carried instructions are considered to be within the scope of this invention. As well, the disclosed processes may ascribe various functionalities to elements disclosed herein. However, such allocations of functionalities are presented only by way of example and, in other embodiments, those functionalities can be allocated differently. Thus, it should be understood that the scope of the invention is not limited to any particular allocation or performance of functions. Moreover, while some example methods and processes are concerned with the operation of vehicles in a physical environment, the scope of the invention is not so limited. Rather, such methods and processes can be used in any application or environment where visual recognition of a physical environment is needed and/or where one or more actions are identified and taken based upon such visual recognition.

The logistics of the operation require selected Spherical Functions corresponding to selected degrees l and orders m to be preloaded into a database and made available to the processor or processors in RAM on demand during executing a set of commands and the visual image located on virtual Hemisphere might be understood as interacting with the Spherical Harmonics. In more detail, the visual function (image) is expanded into a series of the spherical harmonics. This interaction results in a set of complex numbers referred to as coefficients of the expansion. The spherical harmonics each have two indices, which are integers, l and m. The indices are theoretically infinite in number, but on a practical level, a selection is limited to a relatively small number of the most informative such indices.

In some instances, no more than a single set of spherical harmonics is needed to generate invariants, discussed in more detail below, while in other embodiments, two or more sets of spherical harmonics can be employed. More particularly, a single set of functions, using the North Pole as the center of a Hemisphere on which the image is projected virtually, can be used. In other embodiments, two or more sets of functions, each with different respective North Pole positions, can be employed.

As used herein, the term ‘invariant’ embraces a sum of all members of the expansion of the visual function, or image, into the series of spherical harmonics having the same value of the index l. The value of the invariant is a real number that is invariant, that is, does not change, when the projected image is rotated on the hemispherical surface. Thus, a set of invariants corresponding to all of the available indices l constitutes a multi-dimensional vector, namely an n-tuple, that is invariant to rotations. Put another way, an invariant is a point or vector in an N-dimensional vector space, where the axes are real numbers. Due to the nature of rotational invariants, they will represent details of the projected visual scene regardless of where the details are located on the hemispherical surface.

In general, determination of the invariants is based upon the values of the index l. The determination of invariants can take the form of a multi stage process that involves multiple hierarchical levels of invariant computation. Moreover, the process of determining invariants can be performed in reference to time quanta. That is, determination of a set of invariants can be performed repeatedly in each of a number of different time intervals, for example, to accommodate changes to elements of a visual scene, or to the entire visual scene itself, that might be experienced as a device moves in a physical environment. Thus, the initial determination of invariants for each of the elements of the projected image can be the first stage of a multi stage process.

Similar processes can be performed for successively higher levels to generate invariants for larger and larger portions of the projected image. All of the levels of an invariant determination process can be performed within a single slice or interval of time. Thus, while multiple invariant determinations on various respective levels can be performed, those multiple determinations may be considered as constituting a single process performed within a particular time interval. Successive and/or prior processes performed in other time intervals can similarly include multiple determinations at various levels.

Over time then, as invariants continue to be generated for the projected image, invariants that meaningfully correspond to the physical object or environment to which the projected image corresponds will be reinforced and retained as being representative of the physical object or environment, while invariants that lack such correspondence will not be reinforced and will eventually be automatically deleted and/or marked as lacking significance. In some embodiments, these designations of invariants can be implemented in connection with a measurement of the frequency with which an invariant recurs. Thus, regularly recurring invariants will be retained, while irregularly or rarely occurring invariants may be rejected.

D. ASPECTS OF FURTHER EMBODIMENTS

With attention now to FIGS. 1-14, details are provided concerning some example embodiments for the construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition.

When a functioning visual system observes outside objects, in overwhelming majority of cases, projections of those objects on a photosensitive surface do not appear stationary and constantly shift from one location to another. It is also safe to assume that the most appropriate surface for such input device is a hemisphere. Intuitively, we presume that the projections of the objects while moving remain the same and are thus easily identified. In reality, however, such identification cannot be assumed to be trivial and a mechanism must be described that guarantees proper identification of constantly moving projections.

The problem is properly mathematically solved if a technical system is designed to compute rotational invariants which are used as descriptors of the visual world. The task of computing invariants is greatly simplified if the visual images are projected on a hemisphere. For the lack of a web-camera with hemispheric photosensitive surface, although such systems might be made available in the future, the “projection” is performed virtually in a computer or a server after the image is digitally read from a web-cam. The images then are assumed to be functions in the Hilbert space, and are expressed via linear combinations of special orthonormal (on the sphere) functions, Spherical Harmonics, that serve as an irreducible functional basis in the Hilbert space.

More accurately, one can say that the light intensity functions on the hemisphere representing projections of images from the outside world, always considered behaving mathematically well, are expressed as sums of Spherical Harmonics multiplied by expansion coefficients (see formula [1] below). Among other things, embodiments of the invention embrace a system that is capable to generate rotational invariants of the Spherical Harmonics from scratch, channel the evolution of such invariants in a highly coherent cognitive tree database, and provide seed invariants for dissemination among identical retail computer visual systems. The function (image) on the hemisphere is thus expressed as:
ƒ(∂,φ)=Σl=0 Σm=−l m=lƒl m Y l m(∂,φ).  [1]
where integer index l is the degree and integer m is the order and Yl m(∂,φ) are the Spherical Harmonics:
Y l m(∂,φ)=N l m e imφ P l m(∂,φ).  [2]
Nl m is a normalization factor,
e imφ=cos(mφ)+i sin(mφ)  [3]
is a complex exponent and ƒl m are the expansion coefficients:
ƒl m=∫0 0 π/2ƒ(∂,φ)Y l m(∂,φ)sin ∂d∂dφ  [4]
and P l m(cos ∂)  [5]
is Associated Legendre Polynomial (ALP). In formula [4] the upper limit of integration over angle ∂ is π/2, not π because we operate on a Hemisphere. Small lateral movements of the elements of the visual scenes (translations) will project on the Hemisphere as rotations with sufficient accuracy if they are not excessive.

In reality, the summation over the degree l (formula [1]) does not go to Infinity but is limited by some large number N and it also does not start from 0 for our purposes but from some number n<N. Coefficients with negative order m have a simple relation with those with positive order m and therefore their time consuming calculation can be avoided.

Also as will be seen, it is possible to represent a visual image with sufficient accuracy while the system ignores some degrees n≤l≤N as well as some orders m, which is important to speed up computations. In some calculations, the angles ∂ and φ will also be range limited. It is important to note that this limited selection is not dynamic. That is, it is performed once and for all while constructing a prototype, and appears static in nature. In embodiments of the invention, the set of indexes l and m are preloaded in the computer memory before the start of the operations however the resulting selection of the indexes differs in different hemispherical zones (see FIGS. 6 & 7).

The nature of the Visual Image on the Hemisphere impacts the choice of individual degrees l and orders m needed for sufficiently accurate description of visual scenes. If the Image is flat and there are few details, or in other words the gradients across the Image are small, most of the coefficients with higher degrees l will end up negligibly small. In order to describe typical images available in our surroundings, with a plethora of details and good contrasts, degrees l in the vicinity and often in excess of 600 are needed in the area around the North Pole of the Hemisphere (see FIG. 12). The period of the trigonometric functions important for discrimination in the azimuthal direction (see formula [3]) are calculated with this formula,
T=/m  [6]
and the “waves” of Associated Legendre Polynomials important for discrimination along the meridional direction are:
2π/(l+0.5)  [7]

Note that in the latter case, the wavelength of ALP does not depend on order m. The optimal situation presents when the dimensions of details to be described on the Hemisphere match the period T in azimuthal direction [6] and the wavelength of the Legendre Polynomials in meridional direction [7].

FIG. 1 discloses aspects of the principle of some example embodiments of the invention. At stage (1) the Image is available physically on the flat surface of a Web-cam or a photo-camera with or without a zoom lens. It is read from there and recalculated in such a way that in the virtual space of the computer or server, it is projected on the Northern Hemisphere of the 2-sphere. The North Pole of the Hemisphere and the Equator are shown symbolically on stage (2) of FIG. 1. Point N (∂=0) is both the Center of the whole Visual Field as well as the North Pole for the set of the Spherical Harmonics defined on it. An identical system can be built using the opposite, Southern Hemisphere, however herein we will be talking about Northern Hemisphere as a representative of both for convenience only.

FIG. 2 shows the relationship between the web-cam and the Northern Hemisphere of the virtual 2-sphere in a 3-D projection mode. It is assumed that the whole process of image fragmentation, calculating the invariants, image recognition and library storage, if needed, takes place within a time quantum e.g. 100 milliseconds and takes place in a multiprocessor computer or server with parallel computing capabilities. The image is fragmented in the virtual space of the computer or server according to orientation and positions of zeros of Spherical Harmonics.

A diagram of Spherical Harmonics in a form of inverse pyramid is shown on FIG. 3, inverse pyramid of ALPs of lower indexes l is shown on FIG. 4 and corresponding Expansion Coefficients of a hypothetical visual function on FIG. 5. FIG. 5 also shows the formula to calculate invariants to rotational transformations which are scalar products.
J lm=−l m=lƒl mƒl m*  [8]
where ƒl m* is the complex conjugate of ƒl m.

If we limit ourselves to a certain selection of degrees n≤l≤N, we will end up with

Figure US10062144-20180828-P00001
=N−n+1 invariants. We will adopt a language that each invariant is marked by a corresponding index l. These invariants are real numbers and one can say they are one-dimensional invariants. The collection of
Figure US10062144-20180828-P00001
invariants might be considered one multi-dimensional invariant, or multi-invariant, which is a vector in Euclidean real-dimensional space
Figure US10062144-20180828-P00002
, and they will be used for image characterization and recognition.

If some non-informative degrees l are dropped from the set of

Figure US10062144-20180828-P00001
, then the dimensionality of the multi-invariant will be correspondingly reduced. It is important to note that the vision system computes the expansion coefficients (see formula [4]) but never restores the original image (see formula [1]). Instead, the vision system computes rotational invariants using formula [8]. The computed invariants are being used to represent the images in question.

For the description of realistic visual images much higher degrees l are needed. Examining such diagrams helps in determination of the set of Spherical Harmonics that are practically useful for describing outside visual world. Some example diagrams are addressed immediately below.

First, we need to take into account the distribution of visual acuity over the 2-hemisphere. FIG. 6 shows the North Hemisphere of the 2-sphere with two letters H projected and a set of meridians and parallels. The latter are drawn every 10°. Thus the angular distance between the North Pole and parallel number 2 is 20°. It is clear that the center of the virtual Hemisphere around the North Pole will have to deal with smaller details and the periphery with larger angular details. We've divided the Northern Hemisphere into 4 areas: {(0°, 20°), (20°,40°), (40°,60°), and (60°, 90°)}. The Central Zone should have components of the Spherical Harmonics (complex exponent in azimuthal direction and ALPs in meridional direction) with such degrees l and orders m that will provide maximum resolution. The Para-central Zone will need both functions with lower degrees and orders (indexes). And the Peripheral Zone will need even lower indexes.

At least some example embodiments of the invention also presuppose that the whole virtual Northern Hemisphere of the 2-sphere is divided into small rectangles (tesserae) shown on FIG. 7 and the calculation of the Expansion Coefficients of visual images and corresponding invariants takes place first in such rectangles. The size of such spherical rectangles varies with the distance from the North Pole. They are the smallest in Central Zone, larger in Para-central Zone and even larger in Peripheral Zone.

As shown in FIG. 1, the system examines the “smallest” invariants by trying to find a match for them in the Library of Invariants (stage 4) and if no match is found these invariants are stored in the Library at the corresponding level. Then at the next stage the invariants calculated in two adjacent smallest rectangles are added together (stage 5 and 6). Mathematically the additivity can be shown as follows.

Particularly, assume that we have experimental light intensity (visual) function ƒ(∂, φ) defined on two adjacent smallest rectangles. We can also assume that we can express experimental light intensity functions in each adjacent smallest rectangles as ƒ1(∂, φ) and ƒ2(∂, φ). It will also be assumed, for the purpose of this particular demonstration, that the visual function outside the area of combined two smallest rectangles is zero. Each of the above three functions can be expressed through their corresponding expansion coefficients as:
ƒ1(∂,φ)=Σl=0 Σm=−l m=lƒl m Y l m(∂,φ)  [9]
ƒ2(∂,φ)=Σl=0 Σm=−l m=l g l m Y l m(∂,φ)  [10]
ƒ(∂,φ)=Σl=0 Σm=−l m=l q l m Y l m(∂,φ)  [11]

We then fix one particular degree

Figure US10062144-20180828-P00003
, which defines a subspace in the Hilbert Space of Spherical Harmonics, and compute invariants in two adjacent smallest rectangles, and their union:
J 1=
Figure US10062144-20180828-P00004
ƒ
Figure US10062144-20180828-P00003
mƒ
Figure US10062144-20180828-P00003
m*  [12]
J 2 =
Figure US10062144-20180828-P00004
g
Figure US10062144-20180828-P00003
m g
Figure US10062144-20180828-P00003
m*  [13]
J=
Figure US10062144-20180828-P00004
q
Figure US10062144-20180828-P00003
m q
Figure US10062144-20180828-P00003
m*  [14]

As a consequence of the linearity of the system and orthogonality of the Spherical Harmonics, the additivity of invariants is preserved:
J=J 1 +J 2  [15]

The boundaries of such rectangles are naturally assumed to coincide with some zeros of the Associated Legendre Polynomials along some parallels and zeros of components of the complex exponent functions, cosine and/or sine trigonometric functions, along some meridians. It is assumed that a few waves of either function are contained in each spherical rectangle. Further details will now be provided concerning some example functions.

FIG. 8 shows the Complex Exponent (formula [3]) with two values of index m, m=16 and m=160. One can trace such functions while moving in the direction of any parallel on 2-sphere with Spherical Harmonics defined on it. It is clear that such functions are defined on each parallel (a full circle) of the Hemisphere and their usefulness for detail discrimination depends on their position on the sphere. If we take an exponent function with index m very close to the North Pole, the waves will be close enough to “respond” to very small details and produce expansion coefficients which are meaningful. If we move away from the Center of the overall visual field (the North Pole), larger indexes m will be needed to accomplish a similar task.

FIG. 9 shows three fully normalized ALPs with the same degree l=50 and different orders m: m=2, m=12, and m=25. One can see that an increase in the value of order m increases the length of the portion of the polynomial that is infinitesimally small as compared with the “regular” values at larger angles. For the polynomial P50 12 this flat area is almost 10°. For the order m=25 this flat area is more than doubled. This means that if we decided to use a polynomial with degree l=50 in the vicinity of the North Pole, in Central Zone, we can neglect orders m>5 because the computed expansion coefficients will be too small anyway. It must be kept in mind that we compute expansion coefficients and corresponding invariants “locally,” in rather small areas on 2-hemisphere, at the first stage.

FIG. 10 shows a collection of six fully normalized Associated Legendre Polynomials which are paired for comparison (P50 2, P50 3), (P50 12, P50 13) and (P50 2, P51 2). One can see that with degree l=50 small variations in order m cause considerable phase shift in ALPs, however, a small variation of degree l does not change the phase significantly in the area around the North Pole (P50 2, P51 2). This latter observation tells us that in a certain angular range some degrees l might be either ignored or the invariant for index l can be used also in lieu of neighboring l+1 or l−1 invariants because we will integrate in small spherical rectangles.

FIG. 11 shows a collection of six fully normalized Associated Legendre Polynomials which are also paired for comparison. The top pair in FIG. 11 is (P100 2, P100 3). We can see that there are waves of these two functions close to the North Pole. The middle two functions (P100 12, P100 13) have degree l=100 and orders m=12 and m=13 for comparison. We can see that higher orders m created flat areas near the North Pole. Therefore, such orders cannot be used for calculation of invariants in this angular area. The lowest two functions have “adjacent” degrees l=100 and l=101 and identical order m=2 (P100 2, P100 2). It can also be seen that in a wide range around the North Pole approximately from 0° to almost 20° which coincides with the Central Zone in our definition, the last two Polynomials are almost identical. This picture suggests that we can skip some degrees l when we enter area of much higher degrees l, because they may produce near identical invariants or alternatively l-marked invariant can be used also in lieu of l−1 and/or l+1 invariants.

Now we will see how a significant increase in the value of the degree l affects the shape of the Polynomials. FIG. 12 compares five polynomials. Order m=2 is identical for all of them. Degrees l are chosen from the set: {200,300,400,500,600}. Those polynomials are fully normalized and their values do not exceed 1.0. We can see that even when the degree l=200, the waves of the Polynomials might be too coarse to describe fine details of the visual image. Higher degrees l are needed. The problem here, however, is that the increase in the degree l causes the flat area to expand significantly and such polynomials, if positioned around the North Pole at the Center of the whole Visual View, will be useless for computation of expansion coefficients of the fragments of the Image located around the North Pole.

This problem can be resolved however by rotating a set of Spherical Harmonics with high valued degrees l away from the North Pole through Tait-Bryan angles α, β, and, γ to expose the central area around the Center of the Visual Field to meaningful wavelength. FIG. 13 shows such a situation, and comparison with the Polynomial P600 2 on FIG. 12 which explains why the shift will be effective, however, it will be appreciated that more than one such shifted set might be needed.

In reference to FIG. 13 and the idea of using shifted (rotated) Spherical Harmonics sets, each set may be considered to be made from the central Spherical Harmonics set after a rotational Operator

Figure US10062144-20180828-P00005
(α, β, γ) is applied with α, β, γ being Tait-Bryan angles, where α is a rotation around the axis X, β about the axis Y and γ around the axis Z (roll, pitch, and yaw). Ref [1], identified below, presents formulas for recalculating the point values Yl m(∂,φ) of the rotated Spherical Harmonics from the “stationary” ones which will allow to recompute the expansion coefficients as well if needed.

Now we will consider again the Flow Chart Diagram of FIG. 1. The Image is read from the flat surface of the Web-cam (stage 1) and programmatically transformed as if it were projected to the Northern Hemisphere of the 2-sphere (stage 2) to take advantage of identifying images with the help of their rotational invariants. The image is then fragmented into small Spherical Rectangles whose boundaries coincide with some zeros of ALPs along spherical parallels, and sine or cosine trigonometric functions along meridians (stages 3 and 4). The positions of all rectangles are fixed. The images are then “expanded,” in all smallest such rectangles, into the sets of Spherical Harmonics with such degrees l and orders m that provide maximum meaningful values for the expansion coefficients. The Euclidean invariants are then computed using the expansion coefficients.

At this stage, the Euclidean invariants which are vectors in n-dimensional vector space with axes that are real numbers are compared with those stored in the Library in search of a match. If there is no match such invariants are stored in the Library with the date stamped (“NO” direction on FIG. 1). If there is a match such invariants might be needed for image recognition at a higher level. The higher level means the second operation whereas the invariants computed in the smallest neighboring rectangles are added together to provide invariants of the larger visual rectangles. The fact of the match is registered and the date stamped (“YES” direction on FIG. 1). Only the last dates are saved, thus new dates are superimposed on the old ones, however, a counter of matches keeps the overall number of matches preserved. Each Invariant (a multi-vector) has a set of associated counters which keeps track of semantic associations between neighboring invariants. For example, a multi-invariant for an index finger would be associated with the thumb and/or the hand.

In either case (match, or no match), these invariants are kept for consolidation at higher levels: 5 and 6. At level 5, the two adjacent rectangles in azimuthal direction are virtually joined together and their union invariants are computed via expression [15]. At the same time two adjacent Spherical Rectangles are joined in the meridional direction (stage 6). Each time the Library is questioned for a possible match. The consolidation continues at the subsequent stages which results in invariants of still larger parts of the visual image.

On a practical level the invariants being multi-vectors in Euclidean vector space,

Figure US10062144-20180828-P00002
, cannot be expected to be matched with absolute precision. Some tolerance is introduced to make the matches approximate. After a certain period of exposure to a particular situation, perhaps a street corner or an airport waiting room, to name a few illustrative examples, the collected invariants may be programmatically uploaded into a similar computer system that has never had any similar exposure. Thus, a retail operation of producing image recognition systems of similar designation may be easily arranged.

Many technical applications will benefit from a competent Binocular Vision system, examples of which are disclosed in Bipolar Spherical Harmonics (BiPoSH) formula [16] and in FIG. 14. It will be understood that the Binocular Vision requires two independent visual channels.

The BiPoSH (from here on, subscript 1 will refer to one visual channel and subscript 2 to the second visual channel) are expressed as:

Figure US10062144-20180828-P00006
l 1 l 2 LM(ξ,θ)≡{Y l 1 (ξ)
Figure US10062144-20180828-P00007
Y l 2 (θ)}LM:=Σm 1 m 2
Figure US10062144-20180828-P00008
l 1 m 1 l 2 m 2 LM Y l 1 m 1 (ξ)Y l 2 m 2 (θ)  [16]
Where:
Figure US10062144-20180828-P00007
is a tensor product in Hilbert Space of Spherical Harmonics
∥ξ∥=∥θ∥=1.0 are two unit vectors on 2-hemisphere(s) and
l 1 ≤m 1 ≤l 1
l 2 ≤m 2 ≤l 2
and
Figure US10062144-20180828-P00008
l 1 m 1 l 2 m 2 LM are Clebsh-Gordan coefficients. Available C++ program may be used to compute these.

The conditions for Clebsh-Gordan coefficients being non-zero:
M=m 1 +m 2 and
l 1 −l 2 ∥≤L≤l 1 +l 2

Bipolar Spherical Harmonics constitute an orthonormal basis on

Figure US10062144-20180828-P00009
2(
Figure US10062144-20180828-P00010
2×
Figure US10062144-20180828-P00010
2) and therefore any function:
ƒ(ξ,θ)ϵ
Figure US10062144-20180828-P00009
2(
Figure US10062144-20180828-P00010
2×
Figure US10062144-20180828-P00010
2),
can be expanded into a series of Bipolar Spherical Harmonics:
ƒ(ξ,θ)=Σl 1 l 2 LM g l 1 l 2 LM
Figure US10062144-20180828-P00006
l 1 l 2 LM(ξ,θ)  [17]
gl 1 l 2 LM are expansion coefficients of binocular image in a series of Bipolar Spherical Harmonics.
g l 1 l 2 LM=
Figure US10062144-20180828-P00011
ƒ(ξ,θ)
Figure US10062144-20180828-P00006
l 1 l 2 LM(ξ,θ)dξdθ  [18]
g l 1 l 2 LM=
Figure US10062144-20180828-P00012
l 1 l 2 LMm 1 m 2
Figure US10062144-20180828-P00013
al 1 m 1 al 2 m 2 *
Figure US10062144-20180828-P00014
(−1)m 2
Figure US10062144-20180828-P00001
l 1 m 1 l 2 −m 2 LM  [19]
Where:

  • al 1 m 1 is an expansion coefficient in a series of Spherical Harmonics for monocular vision from left or right channel, and
  • Figure US10062144-20180828-P00013
    al 1 m 1 al 2 m 2 *
    Figure US10062144-20180828-P00014
    is an inner product in the functional (Hilbert) vector space of Spherical Harmonics.

The quantity we need, is here:

Figure US10062144-20180828-P00015
LM=−L M=L
Figure US10062144-20180828-P00012
l 1 l 2 LM
Figure US10062144-20180828-P00012
l 1 l 2 LM  [20]

It is a one-dimensional Bipolar Rotational Invariant marked by the index L. Construction of the multi-invariants for binocular vision involves using two web-cams or two photo-cameras separated by a known distance (d) and pointed in approximately the same point in space. The computed invariants will follow this protocol.

First, monocular expansion coefficients will be computed. Then, simultaneously invariants will be computed for each monocular side and the binocular invariants according to formulas [19] and [20]. The monocular invariants will be handled by one Librarian and will be stored in one Library, but the binocular invariants will need their special Library with connections preserved between the binocular invariants and corresponding monocular ones. Some technical systems may employ three channels working in concert. The mathematical description for such systems will be Tripolar Spherical Harmonics (see Reference [1]) which are constructed analogously.

Distance estimation to the outside objects can be easily done with the described vision system. First, a rough estimation can be done by taking into account how many spherical rectangles it took to construct a particular rotational invariant. The second, more accurate, method is to calculate the vectorial difference between multi-invariants for the left and right visual monocular channels. When the viewed object is remote, the difference must be small, when the object is close, the difference will increase proportionally. And, thirdly, a comparison of some components of binocular invariants with corresponding monocular ones from both channels will afford another estimate of the distance to the objects.

The following references referred to in this disclosure are incorporated herein in their respective entireties by this reference: [1] D. A. Varshalovich, A. N. Moskalev, V. K. Khersonskii, Quantum Theory of Angular Momentum, World Scientific Publishing Co, Pte. Ltd 1989; [2] Sergey Kazantsev, Singular value decomposition for the cone-beam transform in the ball, Sobolev Institute of Mathematics, Novosibirsk, Russia; and, [3] Nidhi Joshi, Aditya Rotti, Tarun Souradeep, Statistics of Bipolar Representation of CMB maps, Centre for Theoretical Physics, India.

E. EXAMPLE COMPUTING DEVICES AND ASSOCIATED MEDIA

The embodiments disclosed herein may include the use of a special purpose or general-purpose computer or a server including various computer hardware or software modules, as discussed in greater detail below. A computer or a server may include a processor or multiple processors thus affording parallel computing capabilities, and computer storage media carrying instructions that, when executed by the processor and/or caused to be executed by the processor, perform any one or more of the methods disclosed herein. An external database system exemplified by Microsoft SQL Server or Oracle Database or, alternatively, a simplified database system designed specifically for the processes disclosed herein may be part of one of the embodiments of the invention.

As indicated above, embodiments within the scope of the present invention also include computer storage media, which are physical media for carrying or having computer-executable instructions or data structures stored thereon. Such computer storage media can be any available physical media that can be accessed by a general purpose or special purpose computer provided this storage media can operate at adequate read/write speeds.

By way of example, and not limitation, such computer storage media can comprise hardware such as solid state disk (SSD), RAM, ROM, EEPROM, CD-ROM, flash memory, phase-change memory (“PCM”), or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other hardware storage devices which can be used to store program code in the form of computer-executable instructions or data structures, which can be accessed and executed by a general-purpose or special-purpose computer system or a server to implement the disclosed functionality of the invention. Combinations of the above should also be included within the scope of computer storage media. Such media are also examples of non-transitory storage media, and non-transitory storage media also embraces cloud-based storage systems and structures, although the scope of the invention is not limited to these examples of non-transitory storage media.

Computer-executable instructions comprise, for example, instructions and data which cause a general purpose computer, special purpose computer, or a server, or special purpose processing device to perform a certain function or group of functions. Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts disclosed herein are disclosed as example forms of implementing the claims.

As used herein, the term ‘module’ or ‘component’ can refer to software objects or routines that execute on the computing system. The different components, modules, engines, and services described herein may be implemented as objects or processes that execute on the computing system, for example, as separate threads. While the system and methods described herein can be implemented in software, implementations in hardware or a combination of software and hardware are also possible and contemplated. In the present disclosure, a ‘computing entity’ may be any computing system as previously defined herein, or any module or combination of modules running on a computing system.

In at least some instances, a hardware processor is provided that is operable to carry out executable instructions for performing a method or process, such as the methods and processes disclosed herein. The hardware processor may or may not comprise an element of other hardware, such as the computing devices and systems disclosed herein. In terms of computing environments, embodiments of the invention can be performed in client-server environments, whether network or local environments, or in any other suitable environment. Suitable operating environments for at least some embodiments of the invention include network computing environments.

The present invention may be embodied in other specific forms without departing from its spirit or essential characteristics. The described embodiments are to be considered in all respects only as illustrative and not restrictive. The scope of the invention is, therefore, indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.

Claims (17)

What is claimed is:
1. A method of performing electronic visual image recognition, the method comprising:
computing local invariants of fragments of a visual image obtained by a camera in each of a plurality of small spherical rectangular regions on a virtual hemisphere;
storing the local invariants in a library of invariants if the local invariants are not already present in the library of invariants;
adding two local invariants from two neighboring rectangular areas to create a combined vector multi-invariant and determine if the combined vector multi-invariant is an existing multi-invariant previously stored in the library of invariants; and
comparing the combined vector multi-invariant to the library of invariants to recognize a physical object in the visual image.
2. The method as recited in claim 1, wherein the visual image is an image of a physical environment.
3. The method as recited in claim 1, further comprising discarding local invariants that do not correspond through regular recurrence with a physical environment or physical object on which the visual image is based.
4. A non-transitory storage medium having stored therein computer-executable instructions which, when executed by one or more hardware processors, implement and/or cause the implementation of the method of claim 1.
5. The method as recited in claim 1, further comprising adding local invariants and combined vector multi-invariants of small and larger areas in various combinations to provide description of the visual image with still larger multi-invariants from larger areas that are unions of the smaller areas on the virtual hemisphere; and wherein comparing the combined vector multi-invariant to the library of invariants to recognize a physical object in the visual image comprises comparing a larger multi-invariant from a larger area to the library of invariants.
6. The method as recited in claim 1, wherein the library of invariants comprises a library of invariants generated in a first system which is electronically provided to a second system which computes the local invariants of fragments of a visual image obtained by a camera in each of a plurality of small spherical rectangular regions on a virtual hemisphere, wherein the camera is in operative communication with the second system.
7. The method as recited in claim 1, wherein computing local invariants of fragments of a visual image obtained by a camera in each of a plurality of small spherical rectangular regions on a virtual hemisphere comprises
gathering data concerning the visual image; projecting the visual image on a virtual hemispherical surface and processing the visual image in the span of a time quantum of one possible value of 100 milliseconds, with a system to be made available immediately to resume processing the next visual image and calculating local invariants in the next time quantum in case of either static or dynamic environment it is facing;
defining a small number of circular zones on the virtual hemispherical surface each characterized by a distance from the center of the virtual hemispherical surface, expected visual acuity and special selection of subsets of indexes defining Spherical Harmonics in each zone,
defining a preloaded set of spherical harmonics corresponding to expected positions of the visual images on the virtual hemispherical surface and expected discrimination in the circular zones of the expected positions where the visual images are located;
selecting ‘l’ and ‘m’ indexes;
fragmenting the spherical harmonics into a set of rectangular elements and associating an ‘l’ and ‘m’ indexes with each element; and
providing operational opportunity for computing local invariants for each of the elements, using ‘l’ and ‘m’ values (l-marked invariants); and
combining l-marked invariants into one multi-invariant.
8. The method as recited in claim 7, wherein the local invariants are determined using no more than a single set of spherical harmonics.
9. The method as recited in claim 7, further comprising determining additional invariants, wherein one or more of the additional invariants is a sum of two or more combined vector multi-invariants previously computed within smaller hemispherical areas during the same time quantum.
10. The method as recited in claim 9, wherein comparing the combined vector multi-invariant to the library of invariants to recognize a physical object in the visual image comprises comparing the determined additional invariants.
11. The method as recited in claim 7, wherein one or more two-dimensional spherical elements resulting from the fragmentation of spherical harmonics has a rectangular shape, and some elements are larger than other elements.
12. The method as recited in claim 7, wherein the set of spherical harmonics has an associated North Pole that is unique to that set and serving as a reference point.
13. The method as recited in claim 12, wherein the l index is higher for an element that is closer to the North Pole, and the l index is lower for an element that is farther from the North Pole.
14. The method as recited in claim 7, wherein the set of elements covers the virtual hemispherical surface.
15. The method as recited in claim 7, further comprising defining an additional set of spherical harmonics.
16. The method as recited in claim 1, further comprising taking, or not taking, one or more actions based upon results of the comparison of the combined vector multi-invariant to a library of invariants.
17. A physical vehicle configured to operate in a physical environment that is either static or in motion, wherein the visual image is an image of the physical environment, and wherein the physical vehicle is a land-based vehicle, an amphibious land/sea vehicle, or an airborne autonomous or human controlled vehicle which is configured to perform the method of claim 1 and to take, or not to take, one or more actions based upon results of the comparison of the combined vector multi-invariant to a library of invariants comprises controlling an action of the physical vehicle.
US15642836 2016-08-30 2017-07-06 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition Active 2036-09-29 US10062144B2 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US15251751 US9727945B1 (en) 2016-08-30 2016-08-30 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition
US15642836 US10062144B2 (en) 2016-08-30 2017-07-06 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US15642836 US10062144B2 (en) 2016-08-30 2017-07-06 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition
US15709954 US9858638B1 (en) 2016-08-30 2017-09-20 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
US15251751 Division US9727945B1 (en) 2016-08-30 2016-08-30 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15709954 Continuation-In-Part US9858638B1 (en) 2016-08-30 2017-09-20 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Publications (2)

Publication Number Publication Date
US20180060999A1 true US20180060999A1 (en) 2018-03-01
US10062144B2 true US10062144B2 (en) 2018-08-28

Family

ID=59410750

Family Applications (2)

Application Number Title Priority Date Filing Date
US15251751 Active US9727945B1 (en) 2016-08-30 2016-08-30 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition
US15642836 Active 2036-09-29 US10062144B2 (en) 2016-08-30 2017-07-06 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Family Applications Before (1)

Application Number Title Priority Date Filing Date
US15251751 Active US9727945B1 (en) 2016-08-30 2016-08-30 Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Country Status (1)

Country Link
US (2) US9727945B1 (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9727945B1 (en) 2016-08-30 2017-08-08 Alex Simon Blaivas Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition
US9858638B1 (en) 2016-08-30 2018-01-02 Alex Simon Blaivas Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6083353A (en) 1996-09-06 2000-07-04 University Of Florida Handheld portable digital geographic data manager
US6486908B1 (en) 1998-05-27 2002-11-26 Industrial Technology Research Institute Image-based method and system for building spherical panoramas
US20040062419A1 (en) 2002-10-01 2004-04-01 Samsung Electronics Co., Ltd. Landmark, apparatus, and method for effectively determining position of autonomous vehicles
US20040101199A1 (en) * 2002-11-22 2004-05-27 Lee Shih-Jong J. Fast invariant pattern search
US6853745B1 (en) 2000-11-03 2005-02-08 Nec Laboratories America, Inc. Lambertian reflectance and linear subspaces
US20060114252A1 (en) 2004-11-29 2006-06-01 Karthik Ramani Methods for retrieving shapes and drawings
US7058217B2 (en) 2001-09-28 2006-06-06 Nec Laboratories America, Inc. Broadened-specular reflection and linear subspaces for object recognition
US7236630B2 (en) 2004-07-26 2007-06-26 Matsushita Electric Industrial Co., Ltd. Image processing method, image processing apparatus, and image processing program
US20100046704A1 (en) 2008-08-25 2010-02-25 Telesecurity Sciences, Inc. Method and system for electronic inspection of baggage and cargo
US20110306413A1 (en) 2010-06-03 2011-12-15 D Young & Co Llp Entertainment device and entertainment methods
US8670621B1 (en) 2012-11-06 2014-03-11 Alex Simon Blaivas Using invariants of spherical harmonics to rotational and translational transformations for image recognition in industrial and security applications
US20140152566A1 (en) 2012-12-05 2014-06-05 Brent A. Safer Apparatus and methods for image/sensory processing to control computer operations
US9002098B1 (en) * 2012-01-25 2015-04-07 Hrl Laboratories, Llc Robotic visual perception system
US20160335790A1 (en) * 2015-05-13 2016-11-17 Intel Corporation Iterative closest point technique based on a solution of inverse kinematics problem
US9727945B1 (en) 2016-08-30 2017-08-08 Alex Simon Blaivas Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6083353A (en) 1996-09-06 2000-07-04 University Of Florida Handheld portable digital geographic data manager
US6486908B1 (en) 1998-05-27 2002-11-26 Industrial Technology Research Institute Image-based method and system for building spherical panoramas
US6853745B1 (en) 2000-11-03 2005-02-08 Nec Laboratories America, Inc. Lambertian reflectance and linear subspaces
US7058217B2 (en) 2001-09-28 2006-06-06 Nec Laboratories America, Inc. Broadened-specular reflection and linear subspaces for object recognition
US20040062419A1 (en) 2002-10-01 2004-04-01 Samsung Electronics Co., Ltd. Landmark, apparatus, and method for effectively determining position of autonomous vehicles
US20040101199A1 (en) * 2002-11-22 2004-05-27 Lee Shih-Jong J. Fast invariant pattern search
US7236630B2 (en) 2004-07-26 2007-06-26 Matsushita Electric Industrial Co., Ltd. Image processing method, image processing apparatus, and image processing program
US20060114252A1 (en) 2004-11-29 2006-06-01 Karthik Ramani Methods for retrieving shapes and drawings
US20100046704A1 (en) 2008-08-25 2010-02-25 Telesecurity Sciences, Inc. Method and system for electronic inspection of baggage and cargo
US20110306413A1 (en) 2010-06-03 2011-12-15 D Young & Co Llp Entertainment device and entertainment methods
US9002098B1 (en) * 2012-01-25 2015-04-07 Hrl Laboratories, Llc Robotic visual perception system
US8670621B1 (en) 2012-11-06 2014-03-11 Alex Simon Blaivas Using invariants of spherical harmonics to rotational and translational transformations for image recognition in industrial and security applications
US20140152566A1 (en) 2012-12-05 2014-06-05 Brent A. Safer Apparatus and methods for image/sensory processing to control computer operations
US20160335790A1 (en) * 2015-05-13 2016-11-17 Intel Corporation Iterative closest point technique based on a solution of inverse kinematics problem
US9727945B1 (en) 2016-08-30 2017-08-08 Alex Simon Blaivas Construction and evolution of invariants to rotational and translational transformations for electronic visual image recognition

Non-Patent Citations (18)

* Cited by examiner, † Cited by third party
Title
Bahri, C., Rowe, D., & Draayer, J. (2004). Programs for generating Clebsch-Gordan coefficients of SU(3) in SU(2) and SO(3) bases. Computer Physics Communications, 159(2), 121-143. doi:10.1016/j.cpc.2004.01.005.
Blaivas, A. (1976). Visual analysis: Theory of lie group representations. Mathematical Biosciences, 28(1-2), 45-67. doi:10.1016/0025-5564(76)90094-8.
Blaivas, A. S. (1976). Visual Analysis: Theory of Lie-Group Representations-II. Computations. Institute of Higher Nervous Activity and Neurophysiology, Academy of Sciences, Moscow, USSR, 485, 101-149.
Canterakis, N. (1999) 3D Zernike Moments and Zernike Affine Invariants for 3D Image Analysis and Recognition, Int. 11th Scandinavian Conference on Image Analysis, 85-93.
Holoborodko, P. (2011). Applied Mathematics and Beyond. Numerical Integration. Retrieved Oct. 9, 2017.
Joshi, N., Rotti, A., & Souradeep, T. (2012). Statistics of bipolar representation of CMB maps. Physical Review D, 85 (4), 1-17. doi:10.1103/physrevd.85.043004.
Kazantsev, S. G. (2015). Singular value decomposition for the cone-beam transform in the ball. Journal of Inverse and Ill-posed Problems, 23(2), 173-185. doi:10.1515/jiip-2013-0067.
Mak, L., Grandison, S., & Morris, R. J. (2008). An extension of spherical harmonics to region-based rotationally invariant descriptors for molecular shape description and comparison. Journal of Molecular Graphics and Modelling, 26(7), 1035-1045. doi:10.1016/j.jmgm.2007.08.009.
Notice of Allowance of U.S. Appl. No. 13/621,276 dated Feb. 3, 2014, filed Nov. 6, 2012.
Notice of Allowance of U.S. Appl. No. 13/621,276 dated Oct. 28, 2013, filed Nov. 6, 2012.
Novotni, M., & Klein, R. (2004). Shape retrieval using 3D Zernike descriptors. Computer-Aided Design, 36(11), 1047-1062. doi:10.1016/j.cad.2004.01.005.
Office Action of U.S. Appl. No. 13/621,276 dated Jul. 10, 2013, filed Nov. 6, 2012.
Rowe, D. J., & Bahri, C. (2000). Clebsch-Gordan coefficients of SU(3) in SU(2) and SO(3) bases. Journal of Mathematical Physics, 41(9), 6544-6565. http://dx.doi.org/10.1063/1.1286768.
Szapudi, I. (2004). Wide-Angle Redshift Distortions Revisited. The Astrophysical Journal, 614(1), 51-55. doi:10.1086/423168.
Varshalovich, D. A., Moskalev, A. N., & Khersonskii, V. K. (1988). Quantum Theory of Angular Momentum. World Scientific Publishing Co., Pte. Ltd., 141-144 and 160-164. doi:10.1142/0270.
Wang, H., Li, S., & Wang, Y. (2004). Generalized quotient image. Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004., 1063-6919. doi:10.1109/cvpr.2004.1315205.
Zhang, L., & Samaras, D. (2006). Face recognition from a single training image under arbitrary unknown lighting using spherical harmonics. IEEE Transactions on Pattern Analysis and Machine Intelligence, 28(3), 351-363. doi:10.1109/tpami.2006.53.
Zhang, L., & Samaras, D. (n.d.). Face recognition under variable lighting using harmonic image exemplars. 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003. Proceedings. doi:10.1109/cvpr.2003.1211333.

Also Published As

Publication number Publication date Type
US9727945B1 (en) 2017-08-08 grant
US20180060999A1 (en) 2018-03-01 application

Similar Documents

Publication Publication Date Title
Chaudhuri et al. Depth from defocus: a real aperture imaging approach
Parent et al. Trace inference, curvature consistency, and curve detection
Marr et al. From understanding computation to understanding neural circuitry
Younes Shapes and diffeomorphisms
Shilane et al. Distinctive regions of 3D surfaces
Koenderink What does the occluding contour tell us about solid shape?
Flusser et al. Rotation moment invariants for recognition of symmetric objects
Avidan et al. Trajectory triangulation: 3D reconstruction of moving points from a monocular image sequence
Hildreth Computations underlying the measurement of visual motion
Khotanzad et al. Classification of invariant image representations using a neural network
Grimson et al. On the sensitivity of the Hough transform for object recognition
Stein et al. Structural indexing: Efficient 2d object recognition
Gindi et al. Optical feature extraction via the Radon transform
Feixas et al. A unified information-theoretic framework for viewpoint selection and mesh saliency
US20100033551A1 (en) Content-Aware Wide-Angle Images
Kindratenko On using functions to describe the shape
Prewitt Object enhancement and extraction
Sinha et al. Deep learning 3D shape surfaces using geometry images
US20100086220A1 (en) Image registration using rotation tolerant correlation method
Huang et al. Biologically inspired features for scene classification in video surveillance
Rahtu et al. Affine invariant pattern recognition using multiscale autoconvolution
Yeshurun et al. Shape description with a space-variant sensor: algorithms for scan-path, fusion, and convergence over multiple scans
US6671400B1 (en) Panoramic image navigation system using neural network for correction of image distortion
US20050244059A1 (en) Image processing method for object recognition and dynamic scene understanding
Knill Perception of surface contours and surface shape: from computation to psychophysics